McMaster researchers develop new, publicly accessible AI tool to help scientists find new antibiotics

“We’re doing something a little bit weird,” says Jon Stokes, an assistant professor in McMaster University’s Department of Biochemistry and Biomedical Sciences.  

His lab, under the leadership of graduate student Autumn Arnold, has developed “the ESKAPE Model,” a new AI tool purpose-built to identify new antibiotics. What’s weird, according to Stokes, is not that their new model can help identify novel drug candidates in the blink of an eye; rather, that they are releasing it to the public immediately — and for free.  

Stokes says standard protocol is to write a research paper about the development of the tool and to submit it to relevant academic journals. If journal editors accept the paper, the research — and the tool itself — would undergo rigorous peer-review by other scientists. This process can help with refinement, provide the research team with suggestions, and even flag potential issues.  

But that can all take a year or more, and Stokes isn’t interested in waiting.  

“We’re still working on the paper that describes this work, but we wanted to get the tool out there in the meantime,” he says. “We know it’s not perfect, but even in its imperfect form it can have immediate utility for anyone doing antibiotic discovery research.”  

Stokes and Arnold are crowdsourcing the peer-review process — they want their new tool in the hands of as many researchers as possible so they can get direct feedback from the people who actually use it. This way, they can continuously refine the model to reflect the long-term, evolving demands of the field.  

The researchers admit that their decision to publish the tool like this initially kept them up at night, but they are now resting assured that their expediency may help get new medicines to patients faster.  

“Perhaps it’s to the detriment of my academic career, but, to me, there are more important things than publications and citation metrics,” Stokes says. “Sitting on this until it’s perfect just didn’t feel right. We hope that getting it out there now will enable us to more rapidly receive user feedback and accelerate others’ ability to discover important new drugs.” 

As its name implies, the new tool specifically screens for chemicals that may have therapeutic potential against the ESKAPE pathogens, a globally-recognized list of the world’s most dangerous and drug-resistant bacteria.  

“Due to their ability to resist most antibiotics, there are very few options for treating the infections caused by the ESKAPE pathogens,” Arnold says. “Using machine learning, our tool is allowing researchers from all over the world to rapidly assess their own chemicals for new drug candidates that target these bacteria.”   

Arnold, who developed the tool in collaboration with researchers in McMaster professor Andrew McArthur’s lab, says it was designed to be accessible in more ways than one — not only freely available, but also extremely easy to use. 

“We designed the model so that even people who have no experience with AI can use it with ease,” Arnold says.  

The model simply asks users to input “SMILES codes,” strings of alphanumeric characters that represent the structure of different chemical compounds. Following a simple copy-and-paste, users quickly receive AI-guided predictions about whether or not their chemicals have therapeutic potential against any of the ESKAPE pathogens. 

The research team says ESKAPE Model users can currently screen upwards of 20,000 chemicals in the average workday, without ever picking up a pipette or spending a dollar. For context, Arnold says the same output using a traditional wet lab approach would take several weeks and cost thousands of dollars — and may not ever yield any positive results.  

“This approach allows researchers to prioritize which molecules to test in the lab,” Arnold says. “It can help save people a lot of time and money.” 

Arnold has already used the model to screen 12 million molecules herself — a feat that would take an entire career and then some by traditional means. Importantly, this work has led to the discovery of multiple new antibiotic candidates, which are now under study in the Stokes Lab.   

Stokes, part of McMaster’s Global Nexus and the Michael G. DeGroote Institute for Infectious Disease Research, compares the ESKAPE Model to the Comprehensive Antibiotic Resistance Database (CARD) in that it’s another publicly accessible, McMaster-made tool designed to assist the global research community in the fight against drug-resistant bacteria.  

Given that CARD catalogues genomic data pertaining to drug resistance and the ESKAPE Model can identify new antibiotics to overcome resistance, Stokes says the tandem tools are cementing McMaster’s spot at the cutting-edge of antimicrobial research.  

“McMaster is quickly becoming a one-stop shop for everything antibiotics,” he says.  

The ESKAPE Model officially launched Jan. 27 and is available at eskape.mcmaster.ca.